"Information circulation system" which can analyze and analyze 2 channel TV program live status in Giken Public Relations 2009
It seems that everyone is wearing comments (in essence, less than necessary) such as "Kita ━━━ (゜ ∀ ゜) ━━━ !!" in real time against sports broadcasts, variety shows, TV programs such as animation, etc. It is often seen in the "live board". This act is called "live commentary", for example writing a cheering while watching a relay of professional baseball, inserting Tsukkomi unintentionally while watching various variety programs, if a drama drama is a culprit before a detective I am doing various ways of using it, thinking about it. The content written in there is miscellaneous, but there is no doubt that it is the viewer 's response to the program.
There is no reason for NHK to keep an eye on it, and it seems that "Intelligence Circulation System" is currently considered. It seems to analyze and analyze comments written in real time to make use of future comments on programming and to create a community of viewers.
Details are as below.
It seems that it aims at improving the contact ratio of the net generation
In this system, the advantages of the broadcasting station are "marketing", "improving the contact ratio to the net generation" and "motivation for real-time viewing". It is of course important to understand whether the program is popular or unpopular, but it seems important to be able to see it in real time instead of recording. Also, viewers' benefits include "sharing impressions", "services according to preferences" and "community formation / promotion".
As for comments on TV programs, there is something that you can not divide by just choosing between "good" and "bad", even if it says "interesting" if you follow only that writing I will. Therefore, it seems that conventional evaluation analysis technology can not be used.
The sample is a comment for the big drama "Tenchijin". For example, "Abe-chan Kita ━━━ (゚ ∀ ゚) ━━━ !!" takes only meaning "(Kenshin Uesugi's role)Hiroshi AbeIt was only that it appeared.
When you are live, the scenes reflected on everyone's TV are the same, so even if you omit the subject, unique phrases are used. Besides the "kitter" mentioned above, if a little male is producing a dubious atmosphere, "Ah!"When the pitcher strikes baseball live broadcasting," Spit and strikeout every time "is used.
Therefore, when comment analysis is carried out, first of all, a definite comment of the subject is confirmed, a comment which supplements the subject from the subtitle data of the time when writing was estimated is estimated, and still a procedure of comparing with comments around By stepping on, we will process what is not a comment for the program or not. If the subject is a reliable comment, the correct answer rate is about 96%, and it seems that 6930 out of all 8000 comments are comments on the program contents.
Categorize the unique emotional expression used in writing into 6 patterns of "affirmation" "grief" "contempt" "startle" "negation" "inrush". Analyzing it seems to be difficult because emoticons are symbols only when processed mechanically.
Clustering to group similar comments.
The response speed as a task of the comment input side. In the live performance of "Castle in the sky of castle", there was a load that took the server to fly at the moment when the word "bals" was issued in the scene of the climax before, and in the next laputa live period, There is an episode that the server flew when the Muscat dialogue that "people are like garbage" comes out in the scene before that even though the person was prepared.
The analysis result is planned to be utilized like this.
This is a sample of comments actually attached to the river drama. Apparently it seems that it was the childhood of Naoe Kaneto and Uesugi Shikabin, which was a childhood event, and he was playing Higuchi Rokuro (later Kaneto)Kiyoshiro KatoThere was a comment saying that it was cute.
According to NHK, I have not thought about implementation yet. If the system that analyzes the 2 ch DAT file can do, I think that it will be able to listen to live voice as to whether the program was interesting by analyzing threads of various programs on the live board, but I do not know the reason Is it difficult in the live situation where ASCII art flies?
in Coverage, Posted by logc_nt